View Full Version : "Understanding the Cell Microprocessor" - Anandtech

03-19-05, 06:18 PM

anyone else read this article? very interesting. im very excited about the new consoles even though i favor pc gaming, although, after reading this article i wondered if sony is going to go dig themselves into a hole.

it seamed that they were saying that coding for the cell is going to be very different from coding for the pc or xbox2. to me that means that if there are cross platform ports they will cost a fortune because very little code gets to be re-used.

any thoughts?

03-19-05, 07:14 PM
As a programmer, I'm highly skeptical of the Sony hardware. It appears that (as usual), Sony is focusing on hardware, while MS is focusing on software. Years ago, I used to do lots of low level coding, and enjoyed it. These days, I don't care for lengthly periods of low level optimization, or driver writing. I figure, let the people who made the hardware write the best driver and let me write some cool game to use it. I want hardware designed with the programmer in mind, not some theoretical amazing performance.

Something that can save wacky architecture is good tools. For example, if the machine requires geometry to be split into tiny chunks for processing (like PS2), give me tools to do that as a off line pre process. If the CPU is fussy just to get reasonable performance, give me a good optimizing compiler to do most of the work for me. The problem is, robust flexible tools are a lot of work, and a good optimizing compiler is amazingly difficult to make. Sonys track record for quality tools including compiler is lacking.

The biggest issue I see, is that modern software algorithms that are interesting, such as Physics, Animation, Artificial Intelligence, Dynamic geometry, all require... fast, random access of large amounts of memory. The cell architecture sounds an awful lot like a collection of primitive processors that access tiny amounts of fast memory to do batch processes. This does not lend itself to what I have described as modern, interesting algorithms. It is optimal for things like audio/video encoding/decoding, rendering pipeline, geometry and general data transforms. However, since nVidia is providing a GPU for the PS3, I presume the main cell processors would not be used for these purposes.

Another thing mentioned with cell, is this ability to chain or group sub cpu processing elements together, and also achieve this group processing with remote elements over a network. I don't believe the PS3 is designed to share load with other computers on some kind of network, unless SETI becomes a popular game ;) Any kind of load sharing or processor chaining these days has overhead, either to copy data around, or as delays due to dependencies. At least one of these cell architecture reviews suggested the cell processor is so powerful it could emulate a fast Intel or PowerPC type chip. This doesn't make any sense to me at all.

So that is my opinion, and I hope I'm wrong because there are or will be a lot of PS3 developers expected to produce games to demonstrate this boasted 1000x speed increase and compete with the XBox2.

03-19-05, 07:33 PM
hmm very interesting read greg. from what i have read at m$.com on the new directx and from what others have pieced together about the xbox2, imo m$ seams to be making it very simple to create for both the pc and console (xbox) platform. which, i think, is a good thing for gaming.